Abowd Kramarz Costs Hiring Separation Final

8/8/2019 Abowd Kramarz Costs Hiring Separation Final

1/45

The Costs of Hiring and Separations

John M. Abowd 1 Francis Kramarz 2

January 2003

1 Cornell University, U.S. Bureau of the Census, CREST, NBER and IZA

([email protected])2 CREST-INSEE, CEPR, and IZA ([email protected]). We would like to thankDaron Acemoglu, Josh Angrist, Christel Colin, Emmanuel Duguet, George Jakub-son, Guy Laroque, Philippe Trehorel, and Daniel Verger for helpful comments andsuggestions. We are solely responsible for remaining errors. Abowd acknowledgesthe nancial support of the National Science Foundation (grants SBER 96-18111 andSBR 93-21053, both to the NBER). The data used in this study are con dential butth th i t l i Oth h i t t d i i th d t


2/45

Abstract

In this article, we use the structure of French labor law and the data fromthree linked sources to estimate the costs of hiring, separation, and retirementof employees for a representative sample of French establishments matched witha representative sample of their employees. The estimates are computed using

data from the Wage Structure Survey (ESS), the Workforce Movement Ques-tionnaire (DMMO), and the Occupational Structure Survey (ESE). We showthat the estimated separation costs are increasing and mildly concave functionsof the number of exits and include a very large xed component. Estimatedhiring costs are much lower and those associated with short-term contracts areeff ectively zero. Pro t maximizing French rms should adjust employment pri-marily through hiring changes.

JEL Classi cations: J30, D21

1 Introduction

We begin at the intersection of dynamic labor demand analysis and the study of the rms cost functions (initiated in Oi, 1961). The rms economic problem isthe following. Facing economic shocks, the rm must decide to hire or to termi-nate some workers. To compute the optimal decision, for example the number of terminations, the rm must take into account di ff erent types of costs and bene- ts: past hiring costs, past training costs (both of which are sunk), terminationcosts, total compensation, and marginal productivity. In the United States thearrival of a negative demand shock causes the establishment to immediatelyreduce employment through increased separations (Anderson and Meyer, 1994,Davis and Haltiwanger, 1992). The laido ff workers enter unemployment, theexisting unemployed workers exit at the same rate as before the shock; hence,the rate of unemployment increases (Abowd and Zellner, 1985, Blanchard andKatz, 1997). Evidently, the cheaper form of adjustment in the US is to re theemployees, even though there are no direct estimates of these costs. In France,the arrival of a negative demand shock causes the establishment to immediatelyreduce hiringin particular, to eliminate the hiring of employees on xed du-ration contracts (Abowd, Corbel and Kramarz, 1999). The reduction in hiringrates, slows the exit from unemployment without changing the entry rate and,hence, the unemployment rate increases.

Because the set of French employment responses is less familiar to most read-ers, we summarize the relation between establishment size change (employmentgrowth) and the rates of entry and exit of workers in Figure 1. In AppendixT bl A1 h ddi i l d il f h dj b f l


3/45

of employment mobility in France is the xed duration employment contract,which we discuss in detail below. Since more than 70 percent of all employmenttransactions in 1992, the year of our data analysis, involved short-term contractemployees, we believe that the structure of adjustment costs, and the nature of the associated employment institutions, in France must be substantially di ff er-ent from those of the United States. Furthermore, we believe that the French

institutions may be more typical of European institutions and, therefore, areof considerable scienti c interest because di ff erences in the optimal response todemand shocks arise precisely because of the di ff erences in institutional settings.A reasonable working hypothesis for France, and for much of Europe, is thathiring-related adjustment costs are much lower than separation-related adjust-ment costs. In this article we directly address this hypothesis by measuring andmodeling the costs of adjustment in a cross-section of French establishments.

Once the rm decides to terminate some of its workers or to hire new work-ers, we observe the associated costs directly. We analyze these costs in terms

of both variable and xed adjustment costs using a cost function in which theexplanatory variables are (1) the number of workers that entered or left the rm(variable part) and (2) a xed cost paid whenever some workers were hired orterminated. In addition, for a sample of terminated workers we directly observethe severance payment or the retirement bonus and individual characteristics.The direct payments to the workers di ff er from the total termination or retire-ment costs per separation because of indirect, but measurable, costs associatedwith restrictive French employment laws. Our ability to measure both types of termination costs allows us to examine the selection process of the terminated

workers and the structure of the costs in terms of the individual observablevariables. Finally, we observe the size of the personnel department which givesus a measure of other types of costs induced by human resource management,although we do not have any direct measure of the foregone output componentof adjustment costs.

We estimate the costs of hiring, separation, and retirement for a representa-tive sample of French establishments matched with a representative sample of workers employed in those establishments. The data were collected in 1992. Wecompute establishment-based estimates using French data from three matched

sources. The rst source is the Wage Structure Survey (ESS), which provides theestablishment measures of the hiring and ring costs. It also provides, for someestablishments, the employment and the number of new hires and separations.For units where this information is missing, we match data from the WorkforceMovement Questionnaire (DMMO) which gives, for every establishment with atleast 50 employees, the number of new hires and separations in 1992. Finally,we match data on the establishment personnel department from the Occupa-


4/45

Our results show that the establishment-based adjustment costs of separa-tions are an increasing and mildly concave function of the number of retirementsand an increasing and mildly concave function of the number of terminations.In addition, the xed cost component is a large part of the total adjustmentcosts paid when there are separations. Termination or retirement costs act as xed costs of adjustment for at least two reasons: concavity of the cost function

and the xed cost to be paid when terminations take place. Firms should bunchseparations, not do them gradually. Furthermore, the fact that optimal separa-tions are lumpy implies that the sample of establishments with positive observedseparations will not be representative of all establishments at risk to separateworkers. Our estimated models account for this selection bias. The xed costsof hiring are of much lower magnitude than those for separations. Except forhighly-skilled workers on long-term contracts, there are no variable hiring costs.The working hypothesis is, therefore, con rmed by the data analysis.

We also show that the individual-based estimated retirement or ring costs

consist primarily of a xed component and are only mildly related to the variablepart of the legal formula ( i.e. a proportion of the persons wage that varieslinearly with seniority). These results are consistent with the establishment-level results.

In the next section, we give the salient details of the laws and institutionsof the French labor market. Section 3 summarizes our data. In section 4, wepresent the theoretical and statistical models that motivate our speci cations.The results of the empirical analysis are in section 5. Finally, we conclude andrelate our ndings to the growing interface between empirical labor economics

and the macroeconomics of employment ows and unemployment.

2 Firing: The French Labor Laws

Since 1979, French labor law has recognized two types of regular employmentcontracts: xed duration contracts (CDD, contrat ` a duree determinee), whichcontain a speci ed employment start date, end date and remuneration, but haverestricted use (see Abowd, Corbel, and Kramarz, 1999), and inde nite durationcontracts (CDI, contrat ` a duree indeterminee), which are the normal form of contract and which limit the employers right to terminate the employee asdescribed below. Although their use is formally restricted, CDDs are the morecommon method of hiring.

Since 1982, employment contracts have all been CDI unless the employeeand job qualify for a CDD. Short term employment contracts existed prior tothe legal changes in 1982; however, the designation of a contract type was less


5/45

ment in case of absence, temporary or seasonal demand shock). Such contractsare also used for youth employment programs. Furthermore, selection and test-ing of future permanent employees is allowed under such contracts. The contractcan only be renewed once and its length, including renewal, cannot exceed 18months (24 months for youth employment programs). At the termination of the contract, the worker receives a 6 percent severance payment by law. For

example, in 1992, 80% of all entries into private for-pro t or quasi-public es-tablishments were through CDD. On the other hand, at the same point in timemore than 90% of the stock of employees in these same establishments wereon CDI. For those hired under CDD approximately one in three is eventuallyconverted to CDI (Abowd, Corbel and Kramarz, 1999).

Termination of a CDI is a more complex process. Employer-initiated ter-mination of a CDI employee can take two broad forms ring (licenciement) foreconomic reasons or for cause, and early or normal retirement (preretraite andretraite), both of which are considered terminations under French Labor Law

(30 July 1987).1

Firing for cause under French Labor Law can take two forms ring for serious reason or for very serious misconduct. The latter exemptsthe employer from paying a severance payment, and we have no information onthis type of termination. For all other types of terminations and for retirements,the employer must observe a mandatory notice period and pay a severance pay-ment. An employer can initiate a mandatory retirement of a worker if thatperson is currently eligible to receive the full pension paid by the Social Secu-rity system; that is, if the worker has been employed in a covered job for atleast 37.5 years and is at least 60 years old or if the worker has been employed

less time (the exact amount varies) in a covered job but has reached the age of 65, the mandatory retirement age in most industry-level collective agreements(conventions collectives). Retirement timing is, thus, an employer decision or,at the least, a joint employee-employer decision. The mandatory notice periodfor retirement must be at least as long as that for economic terminations andthe severance payment must also be at least equal to the severance paymentgiven in case of economic termination.

Terminations for economic reasons can a ff ect both individual workers andgroups of workers. All terminated workers bene t from a reemployment priority

within the same rm for one year after the termination date. Valid economictermination reasons include: destruction of the workers job, transformationof the workers job, and major modi cation of the labor contract without achange in the jobleading to termination if the worker refuses to sign the newcontract. Major modi cations of the job occur when, because of bad businessconditions or because of technical change within the rm, existing jobs mustbe re-engineered to t the new circumstances. Technical transformations of the


6/45

is better suited to the re-designed job.Because di ff erent rules apply to individual and collective terminations, we

discuss both in turn. For individual terminations, the employee must be noti- ed in writing of the termination and its justi cation. Although the employerneed not inform the personnel delegate (elected representative of the employeesto the comite dentreprise), the administrative authority (Direction du Travail)

at the Ministry of Labor must be informed. The administrative authority can-not block the termination unless there has been a procedural error. Before thetermination, the employee has the right to an exit interview. If a re-trainingprogram is o ff ered, the details must be given in the exit interview. This proce-dure takes at most 3 weeks. Re-training programs may be tailored to individualneeds and are arranged jointly by the rm, the quasi-governmental agency thatadministers the unemployment insurance system, and the national government.The rm pays 4,500FFr per worker for the re-training program.

There are several types of collective terminations, the rst category distin-

guished by French labor laws is the collective termination of less than 10 workersduring a 30 day period. Most steps in this procedure are similar to those de-scribed above. The employer must consult the personnel delegate or the unionrepresentatives at the rm. The employer must notify the employee and theMinistry of Labor in writing. Each worker has the right to an exit interviewat which the employer may o ff er a re-training program, as described above forindividual terminations.

The second category of collective terminations concerns the dismissal of atleast 10 workers during a 30 day period. The 2 August 1989 law requires that rms with 50 or more employees formulate a social plan before implementinga collective termination of this magnitude or greater. This social plan mustplace a limit on the total number of terminations and lay out plans to facilitatereemployment of terminated workers. The plan may also o ff er a re-training pro-gram, just as the collective terminations noted above did. Union representativesor personnel delegates and the departmental director of the Ministry of Labormust also be informed of the plan. Two public meetings of the works council(comite dentreprise) must be organized with an interval between the meetingsof two to four weeks depending upon the number of terminations proposed. The

works council may require the rm to hire a consulting accountant (at the com-panys expense) to help the council with its analysis. During this period, thedepartmental director of the Ministry of Labor must be continuously informedof the proceedings, the plan, and the names of the proposed terminated workers.The Ministry is responsible for enforcing the procedure but cannot block theterminations if the correct procedures have been followed. The Minister hasone month to con rm the procedural correctness of the collective termination.


7/45

For all terminations, regardless of the number of employees involved, therules governing the mandatory noti cation period are as follows. The noti -cation period is the delay between the workers formal letter announcing thetermination and the actual end of the CDI. Workers with less than 6 monthsseniority are not given notice. For workers with 6 months to 2 years seniority,the notice period is 1 month. The notice period is 2 months for workers with

more than two years of seniority. For engineers, professionals, and managers(cadre), the notice period is three months. If the notice period is not respected,the worker must be fully compensated for the di ff erence between the minimumnotice period and the delay actually experienced in the termination. There are,however, no punitive damages.

Severance payments are calculated as follows. Unless the sector collectivebargaining agreement, the rm-level collective bargaining agreement, or the in-dividual contract specify a more generous formula, the legal minimum severancepayment must be paid to workers with at least two years of seniority. For every

year of seniority at the rm the employer must pay 20 hours if the worker ispaid by the hour or 1/10th of the reference wage if the worker is paid by themonth. The reference wage is computed as the average monthly wage over thelast three months of service at the rm. Furthermore, for most workers, anadditional 1/15 of a second monthly reference wage must be added for everyyear of service beyond 10. This second reference wage is the maximum of the rst reference wage and the average wage over the last twelve months.

When terminated workers would not receive a full-rate retirement pension,early retirement may be an option for the rm in case of the terminations for

economic reasons. Early-retired workers must be at least age 55. The candi-date worker must agree to the early retirement by signing a convention alongwith the employer and the French government. The convention requires that,in consideration of early payment of retirement bene ts, the worker forfeits thediff erence between the minimum severance payment as stated in the sector col-lective bargaining agreement and the legal minimum severance payment. As apart of the early retirement package, the rm must pay a one-time supplementof at least 3% of the daily wage times the number of days the worker would havebeen paid under the collective bargaining agreement until retirement. Actual

supplement rates, again speci ed in the early retirement agreement, lie between6 and 8%. The quasi-governmental agency that manages the unemploymentinsurance system (UNEDIC) pays approximately the same supplement to theearly-retiree. The early retirement payments end as soon as the worker reachesnormal retirement age.


8/45

3.1 The Wage Structure Survey

The rst data set that we use was derived from the most recent wave of theFrench Wage Structure Survey (Enquete Structure des Salaires, ESS 1992).The wage structure survey program, conducted jointly by the French NationalStatistical Institute (INSEE) and the Ministry of Labor, was initiated in 1966by the European Statistical O ffi ce (ESO). However, after the 1966, 1972, and1978 surveys, the ESS was abandoned by the ESO. INSEE decided to resumethis survey given the usefulness and quantity of information collected duringeach wave. The 1992 ESS collects establishment wage information as well asindividual wages (employees sampled within the establishment) for a sampleof establishments in the manufacturing, construction, and (some) service in-dustries. The sampling frame has two stages: at the rst stage, productionunits are sampled; at the second stage, individuals employed at these sampledunits are sampled. More speci cally, the universe to be sampled includes all es-tablishments (manufacturing) or rms (construction and service) with at leastten employees. Agriculture, transportation, telecommunication and the servicessupplied to households are excluded from the scope of the ESS. Insurance com-panies, banks, and all other industries where services are supplied to businessesare within the scope of the survey. The sampling frame is derived from theSIRENE system, a uni ed database recording all existing establishments and rms in France. The sampling rate is strati ed according to the industry, re-gion, and the size of the unit. Sampling rates vary from 1 (certainty) for theunits above 500 employees to 1/48 for units between 10 and 20 employees. 2The ESS provides the sampling frame for our subsequent analyses. The otherdata sources from which we added information to the ESS are exhaustive forestablishments above 20 (ESE, see below) and 50 (DMMO, see below). Thus,establishments in the ESS with 10-20 are excluded from all of our analyses.

Data were collected on (1) the wage-setting policy of the establishment and(2) wages and characteristics of a representative sample of the individuals em-ployed at this establishment in that year. We used most individual-level vari-ables available in the ESS for every sampled worker, namely:

total annual compensation inclusive of all employee- and employer-paidbene ts and bonuses but exclusive of non-wage-bene ts,

rm seniority,

type of contract (CDI or CDD),

number of days of employment in the establishment in 1992,


9/45

age,

nationality (French or non-French),

skill-level (in 4 groups),

bonuses for retirement

severance payments for workers that retired or were red in 1992.

The survey distinguishes eight types of bonuses: xed bonuses; bonusesfor personal (not employment-related) events such as a wedding; compensatingbonuses such as bonuses for job-di fficulty; bonuses based on rm performance;bonuses based on team performance; bonuses based on individual performance;bonuses for rm-speci c and exceptional events; and legally-mandated bonuses(primarily severance payments, retirement bonuses and transportation bonuses).To measure the individual severance payment or retirement bonus, we use the

legally-mandated bonus category. The transportation bonus cannot exceed a fewthousand French Francs in 1992, hence we eliminate all bonuses in the categorywith an amount less than 5,000FFr. To distinguish severance payments fromretirement bonuses, we assume, consistent with the law, that all bonuses paid toworkers older than age 55 are retirement bonuses. We assume that all paymentsto workers below age 55 must be severance payments.

The basic research data les for the ESS contain 15,858 establishments with148,976 interviewed employees in 1992. It is also possible to compute, for most of our establishments, within-establishment statistics on the distribution of wages

and seniority. Hence, we computed the rst decile, the rst quartile, the me-dian, the third quartile and the ninth decile of both the wage and senioritydistributions. Even though we do not know the wage or seniority of every re-tired and separated workers, these percentiles will be another way to capturecomponents of the legally-mandated payments (see section 2). In addition, weuse the following establishment-level variables:

total employment: the average full-time monthly employment during theyear 1992;

total hiring, CDD: the number of employees hired on xed duration, short-term contracts;

total hiring, CDI: the number of employees hired on long-term contracts;

total retirements: the number of employees retiring or taking early retire-ment;


10/45

total terminations (other reasons): the number of employees terminatedfor cause during 1992 and reported separately for two groupsengineers,professionals, and managers (cadre) and all other workers;

total terminations (all reasons): the sum of the two categories of termina-tions de ned above;

retirement costs: the sum of early retirement payments paid directly toemployees and regular retirement compensation paid directly to the em-ployees;

severance payments: legally-mandated separation payments discussed aboveplus any other payment made by the employer at separation;

hiring costs: reported employer expenses on job advertising, search rmfees, and compensation of applicants as distinct from employees, explicitlyexcluding training of newly hired employees.

We also use the following training costs:

training hours: the total number of hours of training paid by the rmwhen trainees were directly compensated by the rm, reported separatelyfor engineers, professionals, and managers and for all other workers;

direct training costs: employer paid training expenditures exclusive of trainee labor costs and inclusive of payroll costs for instructors as well asall other direct material costs, such as the rental of equipment and space;

trainees compensation (young): the direct labor costs (total compensa-tion) for young trainees (stagiaires, apprentis, and others);

trainees compensation (others): all other trainees direct labor costs (totalcompensation).

All costs are reported in 1992 French francs. All compensation costs men-tioned above are inclusive of employer-paid payroll taxes and employee-paidpayroll taxes but exclude employer paid bene ts that are not covered by thepayroll taxes. We divide the total compensated training time by the numberof workers in each of the two skill-groups to get a training time per new hiremeasure. As a measure of the annual full time wage rate, we de ne averagelabor costs per employee as the total wage bill reported in the ESS (inclusiveof all payroll taxes and all employee and employer paid bene ts) by the totalemployment. Finally, we use the ESS business environment variables, asked of the responding manager at every establishment or rm:


11/45

3.2 The Workforce Movement Questionnaire

Our second data source was the Monthly Worker Movement Report (DeclarationMensuelle de Mouvement de Main-dOeuvre, DMMO), which is an adminis-trative record of all worker movements at all establishments with at least 50employees. Although this administrative report was created in the 1970s asa part of the governments monitoring of employee terminations, it was rstcomputerized in 1987 for all of France. Each establishment with at least 50employees must report for each employment movement: (1) the nature of thetransaction (a) hire on a long-term contract (CDI), (b) hire on a short-termcontract (CDD), (c) trial hire, (d) transfer in, (e) transfer out, (f) quit, (g) exitfor military service , (h) exit for sickness or death, (i) end of short-term con-tract, (j) end of trial hire, (k) retirement and early retirement, (l) terminationfor economic reasons, and (m) other terminations including for cause; (2) theskill level of the job involved (two-digit occupational code, CS); and (3) the ageand seniority of the employee involved. For this study, we created a working lein which the data were summed to the annual level by skill group aggregatesfor each establishment. The variables used in our analysis were:

total hiring on CDI is the number of long-term contract hires;

total hiring on CDD is the number of short-term contract hires;

total retirements is the number of regular and early retirements;

total terminations (economic reasons) is the number of terminations foreconomic reasons as de ned in section 2.

The DMMO working le contains information for 38,592 establishments (privateand quasi-public) for 1992.

3.3 The Occupational Structure Survey

Our third data source is the 1992 Occupational Structure Survey (Enquete surla Structure des Emplois, ESE), which is an annual administrative data base

of the detailed occupational structure for all establishments with at least 20employees. All establishments from the private or market-oriented public sector(etablissement public industriel et commercial, EPIC) with at least 20 employeeson December 31, 1991 had to complete a questionnaire. The establishmentreports a description of its occupational structure using a 4-digit standardizedclassi cation of occupations. From this classi cation, we de ned the followingvariables:


12/45

supervisors (personnel department) are all administrative technicians (com-pensation specialists, bene ts specialists, bookkeepers, etc.) and supervi-sors employed in personnel or legal departments;

professionals, managers (personnel department) are all professionals (lawyers,MBAs) and managers (human resource managers, personnel directors,compensation managers, bene ts managers, etc.) employed in personnelor training departments.

These occupational categories constitute all of the directly identi able employeesin an establishments personnel or human resource management department.The basic ESE le contains the number of workers employed in these threeoccupations for 99,904 establishments on December 31, 1991.

It is worth noting that most of the specialized cost and administrative in-formation reported in the ESS is known by the rms because of the legal reg-ulations surrounding the employment relation that we discussed in section 2.The French training laws specify that all rms with ten (twenty in some cases)or more employees must spend a proportion of their wage bill for continuingin-service training (see Delame and Kramarz, 1997). This proportion increasedfrom 1 percent at the beginning of the 1980s to 1.2 percent at the end of thedecade to 1.4 percent in 1992. The ow data on entries and exits must bedeclared in the DMMO. The hiring and ring costs are regulated by the lawsdiscussed above. Finally, most of these costs are subject to special tax treatmentand must, therefore, be accounted separately from other direct costs.

3.4 Creation of the Matched Data File

We matched our three source surveys by establishment. In the matched le,we required the establishment to be in both the Wage Structure Survey (ESS)and in the Occupational Structure Survey (ESE). Some establishments do notappear in the DMMO so that all variables are missing from this source. Therewere 7,905 ESS establishments matched. All matched establishments have ob-servations in both the ESS and the ESE. This eliminates all establishments with10-20 employees from the analysis. We did not require that the establishmenthave a record in the DMMO because we use the DMMO to replace missing datain the ESS and ESE as described below.

For those establishments with no data on total employment from the ESS,we used the available information from the DMMO (average of employmenton January 1 and employment on December 31 of the same year). We adoptedequivalent procedure for the following variables: total hires, total separations for

i d l l ti t if th d t t il bl


13/45

was missing, we imputed the values by skill-levels by multiplying the aggregatevariable by the respective shares in these two skill-levels in the establishment asdeclared in the DMMO. Finally, we used the data on entry by type of contractshort-term (CDD) or long-term (CDI)only for those establishments with non-missing data. Table 1a shows the number of available observations for eachvariable from the rm-level data and Table 1b gives the number of observations

for each variable from the individual-level data. The number of observationsused in the di ff erent regressions is shown in our results section.The issue of the representativeness of our ESS analysis data set can be

addressed by comparing the information in the full sample (15,606 establish-ments) with the data in our analysis sample (7,905 establishments). Averageemployment in the full ESS is 124.1 (standard deviation 484.4) whereas aver-age employment in the matched ESS is 195.8 (standard deviation 595.6). Ourestablishments are larger and more variable, on average, for two reasons. First,establishments with 10-20 employees in the ESS are out of scope for the ESE,

and have been excluded from our analysis. Second, we require con

rming infor-mation on the hiring and separation activity and costs either in the ESS or inthe DMMO. If an establishment reports no hiring activity and no hiring costs inthe ESS and reports no hiring activity in the DMMO, the establishment stays inour sample with a value of zero for these variables. Otherwise, if the establish-ment reports no hiring activity and no hiring costs in the ESS but positive hiringin the DMMO, the establishment is dropped from the analysis le because wedo not have the required cost information. If an establishment reports positivecosts but no hiring activity in the ESS or DMMO, then the establishment is

also dropped from our analysis

le because we do not have the required numberof new hires. Similar rules were applied for the separation data.We consider next a more detailed analysis of the 15,606 establishments in

the full ESS and their disposition after the match with the DMMO. Of theoriginal ESS establishments, 3,886 can be matched to the DMMO and 11,720cannot be matched with the DMMO. Among those that can be matched to theDMMO the average employment is 340 employees. Furthermore, 90% of these rms have at least 50 employees. Recall that it is technically possible that some rms with less than 50 employees respond to the DMMO (for instance, if a multi-

establishment company sends a form for all of its establishments). Among thosethat cannot be matched average employment is 52 with an extremely skeweddistribution since the employment at the establishment at the 90th percentileof this distribution is 49, i.e., just below the threshold for responding to theDMMO. Hence, 90% of the non-matching rms do not have to respond to theDMMO and are, therefore, usually excluded from the sample even though someof them do provide data, as mentioned above. Of the 1,118 establishments with


14/45

those which declare some hiring or separation in the ESS (recall that this ispossible in the ESS questionnaire, in which rms where asked entries and exitseven if they did not respond to the DMMO), roughly 50% also declare theassociated costs.

Another issue, which may explain absence of some large establishments, isthe following. Multi-establishment rms have the opportunity to send a unique

le with either each establishment (the normal procedure) or with all establish-ments aggregated. Discussion with the persons in charge of the ESS sources atINSEE reveal that there is no apparent and systematic pattern to the reporting(INSEE 1997). In addition, of the 10,602 establishments with less than 50 em-ployees and are not in the DMMO, 80% of those that declare some separationsalso declare the associated costs. The equivalent fraction for hiring is muchlower, 30%, but as we will see below this is fully consistent with the estimatedzero hiring costs for short-term contracts (see below) which constitute the bulkof hiring in France (Abowd, Corbel, and Kramarz, 1999). Now, if we focus on

those

rms that are found in the DMMO, we see that 75% of those that declaresome separations also declare the associated costs (the gure for hiring is 40%,again consistently with the estimates that will be produced later).

As noted above, the main problem with using ESS without the informationfrom the DMMO is that rms do not respond to all the questions regardinghiring and separation, which are crucial for assessing the extent of adjustmentcosts. Indeed, of the 15,606 establishments, 1,969 respond to the questionsabout hiring (average hires: 65.66). After adding the hiring information fromthe DMMO we have responses from 4,255 establishments (average hires: 61.8,

see Table 1a). Thus, after matching with the DMMO we have more than twiceas many valid responses but there is no evidence that the rms responding tothe DMMO are substantially di ff erent from those responding to the ESS, exceptas regards their size, which we always use as a control variable. Only 391 rmsin the ESS, responded to the question on total retirements (average retirements:18.03, excluding zeros, standard deviation: 23.84). After augmenting the datawith the DMMO responses, we have 2,084 responses (average retirements: 7.21,excluding zeros, standard deviation: 16.95). Apparently, those rms that re-spond to this question in the ESS have retired more workers. The same appears

to be true of

ring for economic reasons. Only 300

rms in the ESS respondedto the question on ring (average ring: 31.32, excluding zeros, standard de-viation: 63.95). After adding the DMMO data, there are 1,133 rm responses(average ring: 19.80, excluding zeros, standard deviation: 66.01). Finally, thecost of retirement per retiree before adding information from the DMMO is76,931FFr whereas it is 134,011FFr after adding DMMO information (see Table1a). Equivalent gures for the costs of ring per separation are 88,153FFr and


15/45

was designed so that if establishments failed to respond to certain questions theinformation could be retrieved from other administrative data sources such asthe DMMO or ESE, which are exhaustive for establishments above 50 and 20employees, respectively). Matching these diverse sources is thus required for avalid statistical analysis. The sampling frame of the ESS is thus preserved, notcompromised, by the additional data (INSEE 1997). We use establishment size

as a control variable in all analyses because this variable is clearly the primarydeterminant of establishments entering our analysis samples.

4 Statistical Models

To incorporate the various aspects of separation or hiring costs, we consider thefollowing statistical models. First, we write the separation cost function, C f (.),as follows:

C f (f t ) = f, 0 + f, 1 f t + f, 2

2f 2t (1)

if f t , the number of involuntary separations (early retirements, retirements, andterminations), is positive, and C h (.), the hiring cost function, as:

C h (h t ) = h, 0 + h, 1h t + h, 2

2h2t (2)

if h t ,the number of new hires, is positive; and where j,k are the coe fficientsof the cost function for j = f, h and k = 0 , . . . , 2. Institutional costs suchas severance payments imposed by the collective agreements, for example, are

included in the ring costs. Help wanted advertising costs as well as trainingcosts are included in the hiring costs. In both cost functions, we included the xed costs, i, 0 , i = h,f, which are modeled explicitly in our empirical results.

The rm will decide to separate or hire workers only under pro t conditionsthat we summarize as follows:

f t > 0 if ezt + vt > 0 (3)f t = 0 otherwise

and to hire under the conditions:h t > 0 if zt + u t > 0 (4)h t = 0 otherwise.

Therefore, our econometric speci cation amounts to generalized tobit modelswhere separation costs are only observed in the separation regime. For sep-


16/45

report ordinary least squares estimates of the same speci cations of equation(1) for those establishments with positive separations.

For hiring, the estimating equations are also a generalized tobit where hir-ing costs are only observed in the hiring regime. The tobit selection is based onequation (4) and the adjustment cost equation is (2) with observably heteroge-nous coefficients in some models. Tables 9 through 11 analyze hiring decisions

in this framework.The information collected in the ESS and DMMO corresponds to the mon-etary and direct part of hiring or separation costs. Production losses afterseparations, selection and training of new hires, and delays due to advance no-tice periods are not quanti ed in our data sources. These are typically activitiesof the personnel department, at least in larger rms. The direct reports of adjustment costs exclude the regular costs of the personnel department activitiesassociated with adjustment of employment. Most rms face a clear trade-o ff inthis area: (a) the rm can bear all of its adjustment costs at the time of hiring

or separation or (b) the

rm can use its personnel department to smooth outthe explicit part of these costs. On the hiring side, this trade-o ff appears as abalance between conducting search and training activities with the personnel de-partment (thus not reported in our data) or conducting the same activities withan external search agency (thus reported as costs of hiring). On the separationside, the trade-o ff is manifested through the use of the personnel department tohandle some of the outplacement activities (not reported in the data) versus theuse of a consulting rm to prepare the exit plan (thus reported as a separationcost). Thus, the analysis of the structure of the personnel department consti-

tutes one way to assess and control the importance of the adjustment costs thatare not measured in our data sources.We use this argument to justify the inclusion of the variables characteriz-

ing the personnel department structure in the equations determining the hir-ing/separation regime. Accordingly, in the selection equations we use the fol-lowing variables: total employment; all training investmentstraining hours(managers, engineers and professionals), training hours (others), direct trainingcosts, trainees compensation (young), trainees compensation (others); goodbusiness conditions; bad business conditions; average labor costs per employee;

and the employment in the personnel department in three skill levels (cleri-cal workers, technicians, professionals and managers). The costs equations are,therefore, identi ed by the inclusion of these variables in the selection equationbut not in the cost equation.

Once the number of separations, either ring or retirements, is optimallycomputed, the rm must decide which worker to retire or terminate. Thissecond stage is modeled as follows. Consider workers indexed by i employed


17/45

ln wi = x i + j ( i ) + i (5)

where the function j (i) indicates the employing establishment j for individuali .3 Given the optimal number of separations, the selection of the workers to be red or retired from the establishment is modeled as follows:

s i = 1 if x i + ln wi + i + i > 0 (6)s i = 0 otherwise

for all i employed at j (i), where i is the residual of equation (5), and i isa statistical residual for the selection equation. Those workers that are either red or retired, for which s i = 1, get a severance payment or a retirement bonusbi according to the formula:

bi = + ( wi sen i,j ( i ) ) + i (7)where is the xed component that the rm pays to the terminated workers, isthe coefficient that measures the additional severance payment that is collectedby terminated workers for each franc of wage and each year of seniority (asdescribed in the legal formula, see 2), sen i,j ( i ) is seniority in excess of the legalminimum to receive a severance payment, and i is a statistical residual, possiblycorrelated with i . A version of equation (7) in which is decomposed using allindividual observable variables x i is also estimated.

Identi

cation of the various eff

ects in equation (7) comes from the presenceof i in the selection equation and its exclusion from the bonus equation. Theeff ect i , the statistical residual in equation (5), measures the deviation of in-dividual i from the wage that prevails in rm j (i) given his or her individualcharacteristics. The presence of the establishment e ff ect j ( i ) in the wage equa-tion controls for between- rm diff erences. Hence, a positive i measures thedeviation of the pay of individual i from the conditional mean within rm j (i).The selection equation includes the estimated establishment wage e ff ects whilethe bonus equation excludes them. Therefore, in the bonus equation, all e ff ects

are due to the interaction of the wage and seniority, as stipulated in the law.

5 Estimation Results

Table 1a reports the summary statistics for our sample of establishments. Our rst estimate of the di ff erent costs is given in this table. The 1992 retirement


18/45

for economic reasons and for cause (other than very serious misconduct). How-ever, the number of workers terminated for cause reported in the ESS and theDMMO includes both workers who were terminated for serious reasons (withseverance payment) and workers who were terminated for very serious miscon-duct (without severance payment). Hence, we give two measures of the costfor terminations. 4 The rst is the ratio of the termination costs to the num-

ber of workers terminated for economic reasons; in 1992, this ratio is equal to214,828FFr. The second is the ratio of termination costs to the total numberof terminated workers (either for economic reasons or for cause); in 1992, thisratio is equal to 95,531FFr. The rst number gives an upper bound on thetermination cost whereas the second gives a lower bound since the total numberof terminated workers may include terminations for very serious misconducts,which are exempted from severance payments. The hiring cost per hire were5,560FFr. This last gure does not include the training costs. These trainingcosts are also shown in this table. Since the ESS does not directly ask for the

training costs for new hires our reported results are computed as the ratio of thetraining costs to total employment. This assumption probably underestimatesthe training costs for the new hires since the establishment total training costswere divided by total establishment employment rather than by the employeesat risk to be trained. We also give estimates of the average number of workersentering, retiring from, and terminated from the establishment in 1992. Twoversions of these statistics are computed. The rst includes establishments withno entries, retirements, or terminations. The second does not include theseestablishments and, therefore, gives us the average size of the groups entering

or leaving the

rm in a given year. In establishments with positive entry, 64.8workers were hired (70 to 80 percent on short-term contracts, see Abowd, Corbeland Kramarz , 1999). In establishments with positive retirements, the averagesize of the group of retirees is 7.2. Finally, in establishments with positive ter-mination for economic reasons, the average size of the group of workers red foreconomic reasons is 20.0 whereas in establishments with positive termination(for economic reasons or for cause) the average size is 15.0.

Table 1b provides summary statistics for our sample of red or retired work-ers. The mean age of red workers is 38 and their average seniority is 10 years.

The age of retired workers is close to 59 and their average seniority is 22 years5

.Our estimate of the mean severance payment is lower than our estimate of the termination cost from the previous table. The same is true for retirementbonuses. Notice, however, that both termination and retirement costs comprisemore than the bonuses collected by the workers, as is apparent from the laws.

Table 2 reports our results for the determinants of the cost of retirementand early retirement. Columns (1) and (3) are estimated using least squares on


19/45

observations with strictly positive retirements and strictly positive retirementcosts, respectively. Columns (2) and (4) are estimated by maximum likelihood(generalized tobit) using all observations with either positive costs and posi-tive retirements or zero costs and zero retirements. Columns (1) and (2) areestimated on all establishments with at least 50 employees while columns (3)and (4) are estimated on establishments with 20 to 50 employees. As described

above, this size threshold is important in France for all of these questionsinparticular, the trade-o ff between retirement and termination should depend onthe size of the establishment. The least squares and tobit estimates are quitediff erent for the largest establishments. In the least squares estimates, the linearpart is very large, the function is strongly concave, and the intercepta measureof xed costsis large and negative. 6 For the tobit estimates, in which the deci-sion to separate is also modeled, the linear part is small, the function is slightlyconcave, and the xed costs, the intercept, are huge: 579,549FFr. This di ff er-ence between the two estimated equations is consistent with our model. Least

squares should be downward biased; only those establishments with low costs,all other things being equal, should retire workers. All estimated coe fficientsfor both methods, expressed in francs per retirement, are large and statisticallysigni cant. 7

As we noted above, retirement costs are concave in the number of retiredworkers. The marginal cost of retiring N workers is estimated at 27 , 435 176N (in 1992 FFr) in those establishments with at least 50 employees, all other thingsequal. The cost of setting up the collective retirement agreement is estimatedto be 579,549FFr.

Table 3, which reports results for retirement costs with the retirees diff

er-entiated by skill-levels, shows that the termination costs of retirement stemprimarily from retiring engineers, professionals, and managers and not from theretirement of other workers. The cost of retiring workers with other skills seemsto be convex although not precisely estimated in Table 3. Therefore, rmsshould optimally group retirements of their skilled workers (concave adjustmentcosts) and retire low-skill workers gradually (convex adjustment costs). Noticehowever that the low-skill component is also not precisely estimated when com-pared to the previous table. Indeed, the number of establishments that report

6 All our models are quadratic functions of the number of retirements, separations, orhires. Even though the laws seem to imply linear costs, a number of unobserved individualcharacteristics of the retired or separated workers that matter to determine the retirement orseparation costs will be captured by this functional form.

7 Estimates for the probit part of the generalized tobit are shown in Appendix Table A2.Most coe ffi cients in the probit equation are signi cantly di ff erent from zero. Surprisingly, thevariable facing bad business conditions has no impact on the retirement probability. Train-ing investmentshours of training per cadre, hours of training per other type of worker, young


20/45

the information on retirements by skills is lower than in Table 2. 8Table 4a reports the results for the costs of ring workers. Our measure of the

number of terminated workers is based on the number of all terminated workersrather than just the number of terminations for economic reasons. The structureof the table is similar to that of Table 2 and the results have the same avor. 9Fixed costs are huge: 1,138,117FFr. The marginal cost of terminating N workersis estimated as 56 , 299 31.2N . Hence, the separation costs are roughly linear, atleast much less concave than those for retirement. As before, the establishmentsthat actually terminate their workers have lower xed costs which translates intolarger proportional costs (see column (1)). Finally, establishments that have lessthan 50 employees have strongly concave costs, no xed costs and those whichindeed terminate workers do not di ff er from those which do not. Once again,the largest rms should optimally group the terminations. 10

The distinction between collective and individual terminations is an impor-tant element of French employment law. One way to address this distinction,

not measured in the data, is to assume that any

rm that terminates 10 workersor more workers in 1992 uses the collective termination procedure while thosethat terminate less than 10 workers necessarily use the individual terminationprocedure. Results using this speci cation are presented in Table 4b. They aresimilar to the results presented in Table 4a. Most of the costs come from collec-tive terminations. Estimates for individual terminations may appear surprisingin that costs only stem from the xed part. Identi cation may be di fficult inthis case because of a lack of variability in the number of terminations, how-ever, rms that terminate workers, irrespective of the procedure, have much

lower

xed costs than those which do not terminate workers in a given year.Table 5 shows the impact of the skill-level of the terminated workers, dis-tinguishing once more between collective and individual procedures (de ned asabove since the law does distinguish by skill-level). The cost of collective termi-nations is an increasing and concave function for engineers, professionals, andmanagers (cadres) and a linear function for other skills. The xed cost part isonce more diffi cult to estimate precisely due to the smaller number of obser-vations. Individual terminations for engineers, professionals and managers arealso costly but the structure is convex whereas individual terminations for other

workers do not appear to cost anything (but note that the estimates resemblethose obtained in Table 4b, also for individual terminations). Such results oncemore demonstrate that rms should optimally group rings into collective pro-cedures whenever highly skilled workers are involved. 11

8 Appendix Table A2 reports the coe ffi cients for probit selection equation associated withthe generalized tobit models in Table 3.

9 This is also true for the probit results see Appendix Table A2 However the variable


21/45

Table 6 tries to decompose the xed cost component obtained in Table 4busing information from the survey about variables that should a ff ect these costs.The regression includes data on wages and seniority of the workers employed inthe establishment. From the data on individual workers, we have computed per-centiles of the within-establishment distributions and used them as regressors.Indeed, we expect to capture elements of the costs of terminations since thelaw stipulates than the payments have to depend on the wage and the seniorityof the terminated workers. Unfortunately, none of these variables have a clearimpact on the costs. The size of the rm is the only element that seems to be(positively) correlated with the costs of terminating workers.

Table 7 presents estimates of equations (6, probit part) and (7, tobit part)for the costs of ring based on individual-level data. These equations are esti-mated by maximum likelihood ( rst two sets of estimates). The probit equationis estimated using all workers employed in an establishment in which there is atleast one individual observation for which the severance payment is known. Thecontinuous part of the tobit is estimated on all workers below age 55 for whomthe information on the severance payment is present. To assess the magnitudeof the biases involved in selection, we also present least squares estimates of the bonus equation for the sample of terminated workers. The rst set of esti-mates ( rst two columns) includes a xed component, an indicator for collectiveterminations in the establishment (measured as above, since no individual in-formation is available), and a variable part (annual wage times seniority). Allresults are consistent with those presented above: tobit estimates di ff er sub-stantially from least squares estimates, in particular the xed cost componentis much larger when the estimation technique accounts for the bias associatedwith selection of terminated workers by the rms while the variable part is muchsmaller. This is consistent with the French labor laws because this part of thebonus is negotiated. When decomposed using individual characteristics (nexttwo columns), the severance payment (expressed in 1992 FFr) is

severance pay = 56 , 823 + 776 seniority + 555 (age seniority )

+7 , 512 (manager = 1) + 4 , 182 (blue collar = 1)

and all other coe fficients are not signi cantly di ff erent from zero. Notice that theresiduals of the two equations are (strongly) negatively correlated which con- rms the bias discussed above. The selection equation also shows that clerks,blue-collar workers, high-wage workers, females, senior or aged employees, orFrench workers are less likely to be selected while overpaid workers (as mea-sured by i , the residual of the wage equation, 5, i.e. workers that are bettercompensated than workers employed in the same establishment with identical


22/45

age to bene t from pre-retirement or retirement programs, see 2). 12 All equa-tions have the same structure as described for Table 7. Selection biases are largeand the xed component is a major part of the retirement bonus. Decomposedin terms of individual observable variables, the cost is

retirement bonus = 294 , 022 2, 718 seniority

3, 181 (age seniority ) + 45 , 932 (manager = 1)Notice that, for all workers, seniority and age coe fficients are large; otherwise,the workers would not bene t from these retirement programs. As for termi-nation costs, the correlation between the selection and the bonus equations islarge and negative. However, age and seniority as well as i enter positively inthe selection equation. 13

Table 9 presents estimates of the cost of hiring new workers. The tablereports least squares results (column 1) as well as generalized tobit estimates(column 2). As above, the two methods give very di ff erent estimates of the xed cost component while the quadratic part is similar in both columns. No-tice, however, that the number of establishments with no hires in 1992 is quitesmall: 117, the di ff erence between the number of observations in column (2)and in column (1). This may explain the high xed cost of hiring for those 117establishments.

Table 10 reports the structure of hiring costs when we di ff erentiate the hiresby skill-level. The hiring costs are due primarily to entry of engineers, profes-sionals, and managers. Table 11 shows, on a small sub-sample, that the type of

contract matters. Hiring costs are due entirely to entry of highly-skilled work-ers on long-term contracts (CDI). Once again, the cost function is increasingand concave; thus, rms should group their hiring of engineers, professionals, ormanagers.

As a nal measure of the reasonableness of our results we computed a simu-lation of the e ff ect of adjustment costs on total labor costs using results from ourestimated models. The results of this exercise are shown in Table 12. We modelan average-size establishment (196 employees, see Table 1a). The table showsthe distribution of the relevant data for younger (age below 50) and older (age

50 and over) workers for six diff

erent occupation groups. The distribution of em-ployment, completed job duration, and wages by age and occupation is based on21 years of data from the Declarations Annuelles de Donnees Sociales (DADS). 14The column entitled Expected Completed Job Duration is an estimate of the

12 We also estimated the same tobit equation with a selection equation run on all potentialworkers with no changes in the estimated coe ffi cients, albeit more precision in the estimatesof the probit equation


23/45

length of a completed employment spell for a worker in the indicated age andoccupation group that is based on a Weibull hazard model controlling for age,sex, and occupation. The column entitled Average Annual Wage is thefull-time, full-year labor costs per employee, including all compensation, payrolltaxes, (in 1992 French Francs). The column entitled Annualized AdjustmentCost is the annualized adjustment cost of replacing the worker. For youngerworkers the adjustment costs are based on two assumptions: (1) involuntaryterminations are ten percent of all terminations (see Table A1 for stable estab-lishments) and (2) the termination costs are summarized in Table 4a column(2). For older workers the adjustment costs are based on three assumptions:(1) all separations of older workers are employer-initiated retirements; (2) theoptimal grouping of retirements is every three calendar years; 15 and (3) theretirement costs are given in Table 2 column (2). 16 The column showing theratio of adjustment cost to total cost (inclusive of adjustment costs) is a mea-sure of the percentage increase in marginal productivity that a worker in thatrow must deliver in order for the employer to recover the investment associatedwith separation costs. 17 The results show that adjustment costs require youngermanagers, engineers and professionals to be 4.4% more productive than in a fric-tionless world. Older managers must only be 2.8% more productive. These arethe two groups for which separation costs have the smallest e ff ect on the totalcosts of employment. The group where separation costs are most important isall ages of blue collar workers, both skilled and unskilled. For these workers theinvestment represented by separation costs increase total annual labor costs bynine percent.

6 Conclusions

In this article, we present estimates of the structure of retirement, termination,and hiring costs using, for the rst time, representative establishment-level datamatched with individual-level information. These costs are directly reported bythe sampled establishments. We provide estimates of the magnitude of thesecosts as well as statistical summaries of their functional shape given the numberof movements or the characteristics of the workers. It appears that both retire-

data. We based our occupation-speci c estimates on these data because they are much morecomprehensive (1/25 sample of the French labor force) than the ESS data used in the basicestimation.

15 The expected number of retirements per year is slightly above two in the simulation. Theaverage number of retirements, excluding zeros, in our sample is 7.2 (see Table 1a). Thisimplies that the retirements have been lumped into batches processed approximately everythree years.


24/45

ment and termination costs are increasing and mildly concave in the numberof retired or terminated workers. Furthermore, the xed costs are very large,giving the rm an incentive to group exits instead of adjusting gradually. Termi-nation costs are largest for collective terminations as opposed to individual ones.These costs are largest for highly skilled employees. Hiring costs also exhibitthe same structure; concave adjustment costs with a strong xed component.But these hiring costs do not have the same structure for all skill levels. Onlyhires of cadres on long-term contracts (CDI) have an increasing and concaveimpact on the cost. For all other skill levels and types of contract, hiring costsdo not depend upon the number of entries. Thus, for hiring costs, the rmshave an incentive to group the managerial (cadre) hiring but no adjustmentcosts for other hiring. The costs of hiring are much less important in Francethan the costs of separations (retirements and terminations). The structure of the personnel department is also related to all types of entries and exits. Weremind the reader that our estimates are based on a single cross-section of es-tablishments and, thus, may be due to compositional e ff ects rather than anysingle rms cost structure.

Our results show, for the rst time, direct evidence on the shape and struc-ture of rm-level adjustment costs in contrast to the vast amount of indirectevidence based upon estimating dynamic labor demand equations (Hamermesh,1995). In France, at least, adjustment costs display two sources of lumpinessthe xed component of the termination or retirement costs (also present in theindividual-based estimates) and the concave shape of these costs, which mayexplain why rms tend to prefer large adjustments over smaller ones, a featurealso found in Caballero, Engel, and Haltiwanger (1997).

On one hand, wages appear to be rigid in France (see Card, Kramarz, andLemieux, 1996). On the other hand, Abowd, Corbel, and Kramarz (1999) haveshown the existence of a considerable amount of worker turnover in France.Indeed, most of these movements stem from the entry and exit of workers onshort-term contracts (CDD). Since the termination or retirement of workers oninde nite duration contracts (CDI) causes adjustment costs in our estimateswhile the termination of CDD workers does not, the conjunction of rigid wages,high ring costs for workers on CDI, and easy hiring and separation for workerson CDD seems to explain the observed behavior of French rms. In particular,our estimates explain why these rms hire primarily on short term contracts,why they reduce entries in bad times without increasing separations, and whyyoung workers nd it difficult to get a job from unemployment.

All of the microeconomic evidence for France has counterparts in the USthat are not very di ff erent from those observed in France. The turnover levelsare quite similar for the two countries. 18 The levels of severance payments


25/45

workers, but the di ff erence is less strong than originally supposed: roughly 40%of full-time workers employed at medium or large employers are covered byseverance payments while only 20% of those employed at employers with fewerthan 100 employees (see BLS, 1993, 1994, 1995). Survey evidence provides evenhigher severance pay incidence, respectively 90% and 66% (see Lee, Hecht andHarrison in BNA, 1996). The same private survey reports an average maximumseverance of 39 weeks for executives, 32 weeks for exempts, and 30 weeks fornon-exempts. In addition, the experience rating in the UI system increasesthe costs of separations in the U.S. relative to France where there is no suchexperience rating system.

Like minimum wages, hiring and separation costs induce labor market rigidi-ties. However, the link between these labor market rigidities and the high Frenchunemployment rate is di fficult to assess. In particular, as Blanchard and Katz(1997) show, increased ows into and out of unemployment do not necessar-ily imply a higher unemployment rate. Hence, just because there is a strongincentive to reduce adjustment costs and to increase the ows into and out of employment by the use of CDD employment contracts in France, the equilibriumrate of unemployment is not necessarily higher. Our evidence addresses the wayin which adjustment costs interact with economic shocks to a ff ect employment ows but we do not have any direct evidence on the steady state employmentconsequences of the French labor laws.


26/45

References

[1] Abowd J.M. , Corbel P., and F. Kramarz (1999): The Entry and Exit of Workers and the Growth of Employment: An Analysis of French Estab-lishments, Review of Economics and Statistics (May), 170-187.

[2] Abowd, J.M., Kramarz, F. and D. Margolis (1999): High Wage Workers

and High Wage Firms, Econometrica 67 (March), 251-333.[3] Abowd, J.M., and A. Zellner (1985): Estimating Gross Labor Force

Flows, Journal of Business and Economic Statistics 3, 254-283.

[4] Bentolila, S. and G. Saint-Paul (1994): A Model of Labor Demand withLinear Adjustment Costs, Labour Economics, 1, 303-326.

[5] Blanchard, O. and L.F. Katz (1997): What We Know and Do Not KnowAbout the Natural Rate of Unemployment, Journal of Economic Per-

spectives, 11, 1, 51-72.[6] Bureau of Labor Statistics (1993): Employee Bene ts in Medium and Large

Private Establishments , 1991 (Washington: GPO).

[7] Bureau of Labor Statistics (1994): Employee Bene ts in Small Private Es-tablishments , 1992 (Washington: GPO).

[8] Bureau of Labor Statistics (1995): Employee Bene ts in Medium and Large Private Establishments , 1993 (Washington: GPO).

[9] Bureau of National A ff airs (1996): Severance Practices, Bulletin of Man-agement, Datagraph, January 11, 1996.

[10] Caballero, R.J., E.M.R.A. Engel, and J. Haltiwanger (1997): AggregateEmployment Dynamics: Building from Microeconomic Evidence, Ameri-can Economic Review, 87, 1,115-137.

[11] Card, D., Kramarz, F., and T. Lemieux (1996): Changes in the RelativeStructure of Wages and Employment: A Comparison of the United States,

Canada and France, NBER working paper no. 5487.[12] Delame, E. and F. Kramarz (1997): Entreprises et Formation Continue,

Economie et Prevision , 127, 63-82.

[13] Hamermesh, D. (1995): Labor Demand, Princeton University Press.

[14] INSEE (1997) L t d l i d 1992 R lt t d l t


27/45

[16] Oi, W.Y. (1961): Labor as a Quasi-Fixed Factor, Journal of Political Economy , 538-555.

[17] Rotbart, G. (1991): Enquete sur la Structure des Salaires, Insee-Methodes,Insee.


28/45

AppendixAppendix Table A1, which is based on the data analysis from Abowd, Corbel

and Kramarz (1999), shows the rates of entry and exit of worker of di ff erentcontract types. The table distinguishes between CDD and CDI workers. Thethree lines of the table show the rates per 100 employees of growing, shrinkingand stable establishments. Appendix Table A2 presents the selection equationsassociated with the models presented in Tables 2 through 5 of the text.


29/45

VariableNumber of

Observations Mean St. Dev.

Total Employment 7,905 195.8 595.6Total Hiring 4,255 61.8 128.9Total Hiring (excludes zeros) 4,060 64.8 131.2Total Hiring (Eng., Prof., Managers cadres) 4,165 8.0 26.4Total Hiring (Others) 4,479 40.3 103.2Total Retirements 3,844 3.9 13.0Total Retirements (excludes zeros) 2,084 7.2 17.0Total Retirements (Eng., Prof., Managers) 3,777 0.7 3.1Total Retirements (Others) 2,749 3.9 10.9Total Terminations (All Reasons) 3,809 11.1 44.8Total Terminations (Economic Reasons, excludes zeros) 1,133 19.8 66.0Total Terminations (All, excludes zeros) 2,845 15.0 51.3Total Terminations (All, Eng., Prof., Man.) 3,668 1.9 8.4Total Terminations (All, Others) 2,215 11.1 41.3Clerical Workers (Personnel Dept.) 7,905 0.81 6.88Supervisors (Personnel Dept.) 7,905 0.84 5.42Professionals, Managers (Personnel Dept.) 7,905 0.30 3.48Retirement Costs per Retiree 1,487 134,011 1,087,370Termination Costs per Termination (Eco. Reasons) 982 214,828 587,309Termination Costs per Termination (All) 2,027 95,531 233,029Hiring Costs per Hire 1,562 5560 26,240Training Hours per Eng., Prof., Man. 7,353 86.2 2037.2Training Hours per Others 7,341 68.0 2403.6Direct Training Cost per Worker 7,896 3025.1 64686.7Trainees Compensation (Young, per Worker) 7,896 458.4 11997.5Trainees Compensation (Others, per Worker) 7,896 1459.1 29427.5

Good Business Conditions (percent) 7,905 4.2Bad Business Conditions (percent) 7,905 31.4Average Labor Costs 7,896 171,022 676,185Within-Firm Wage Distribution (first decile) 7,666 7.206 0.470Within-Firm Wage Distribution (first quartile) 7,666 7.409 0.547Within-Firm Wage Distribution (median) 7,666 7.901 0.693Within-Firm Wage Distribution (third quartile) 7,666 8.438 0.679Within-Firm Wage Distribution (ninth decile) 7,666 8.716 0.598Within-Firm Seniority Distribution (first decile) 7,662 2.831 4.204

Within-Firm Seniority Distribution (first quartile) 7,662 4.612 5.055Within-Firm Seniority Distribution (median) 7,662 8.965 6.678Within-Firm Seniority Distribution (third quartile) 7,662 14.563 8.222Within-Firm Seniority Distribution (ninth decile) 7,662 18.093 9.157Growing Firm (percent) 7,903 45.8Shrinking Firm (percent) 7,903 21.2

Table 1a: Summary Statistics for the Establishment-Level Variables


30/45

VariableNumber of

Observations Mean St. Dev.

Fired Workers:Managers, Engineers, Professionals (percent) 1,705 16.4Clerks (percent) 1,705 17.9Blue-Collar Workers (percent) 1,705 36.0Technicians, Foremen (percent) 1,705 29.7Male (percent) 1,705 69.5French (percent) 1,705 94.7Age 1,705 37.8 8.4Seniority 1,705 10.1 8.2Severance Payment 1,705 18,732 28,990Wage (10 6 FFr) X seniority 1,705 1.69 1.91Log of Annual Wage 1,705 12.0 0.6Wage Residual i 1,705 0.140 0.369

Retired Workers:Managers, Engineers, Professionals (percent) 404 16.4Clerks (percent) 404 20.8Blue-Collar Workers (percent) 404 36.0

Technicians, Foremen (percent) 404 26.8Male (percent) 404 67.5French (percent) 404 92.8Age 404 59.2 2.5Seniority 404 22.6 11.2Retirement Bonus 404 42,324 67,566Wage (10 6 FFr) X seniority 404 4.26 4.19Log of Annual Wage 404 12.2 0.8Wage Residual

i 404 0.272 0.521

Sources: 1a ESS 1992, ESE 1992, DMMO 1992. 1b ESS 1992.

Table 1b: Summary Statistics for Individual-Level Variables


31/45

Mean Coef. Coef. Coef. Coef.Independent Variable (St. Dev.) (St. Err.) (St. Err.) (St. Err.) (St. Err.)

(1) (2) (3) (4)Retirement Costs 956,040 dep. dep. dep. dep.(5,263,787)

Total Retirements 7.72 215,611 27,435 47,729 48,396(16.0) (17,988) (3,468) (15,896) (16,375)

Total Retirements (squared) 314.7 -782.7 -88.0 -2,887.3 -3,045.3(1,860.4) (153.7) (28.7) (987.3) (1,030.1)

Intercept -464,798 579,549 14,406 80,022(169,138) (33,242) (29,761) (58,022)

Number of Obs. 1,370 2,554 116 374R 2 0.179 0.075Log-likelihood -14,688.20 -915.40Sources: ESS 1992, ESE 1992, DMMO 1992.Notes: Columns (1 ) and (3) give least squares estimates; columns (2) and (4) are estimated by maximum-likelihood (generalized tobit model). See Appendix Table A2 for the coefficients of the probit selectionequations. Columns (1) and (3) rely on establishments with strictly positive costs and strictly positiveretirements. Columns (1) and (2) give estimates for establishments with 50 or more employees. Columns

(3) and (4) give estimates for establishments with 20-49 employees.

Table 2: The Cost of Retirement and Early Retirement in 1992


32/45

Mean Coef.Independent Variable (St. Dev.) (St. Err.)

Retirement Costs 1,069,506 dep.(5,997,983)

Total Retirements Eng., Prof., Manag. 1.45 527,761(4.78) (45,489)

Total Retirements Eng., Prof., Manag. (squared) 24.9 -6281.6(269.3) (651.3)

Intercept Eng., Prof., Manag. -140,814(511,498)

Total Retirements Other Skills 6.91 16,633(12.81) (21,284)

Total Retirements Other Skills (squared) 211.8 48.5(976.6) (238.7)

Intercept Other Skills 439,008(538,187)

Number of Obs. 1,414Log-Likelihood -8,080.80

Sources: ESS 1992, ESE 1992, DMMO 1992.

Table 3: The Cost of Retirement and Early Retirement by Skill-Levels in 1992

Notes : Maximum likelihood estimates on all establishments (generalized tobit).


33/45

Means Coef. Coef. Coef. Coef.Independent Variable (St.Dev.) (St.Err.) (St.Err.) (St.Err.) (St.Err.)

(1) (2) (3) (4)Termination Costs 1,619,246 dep. dep. dep. dep.(6,150,543)

Terminations 17.7 92,086 56,299 40,078 51,660(57.0) (3,792) (2,771) (9,364) (12,300)

Terminations 3,565 -18.4 -15.6 -662.7 -1110.4(squared) (52,417) (4.12) (3.04) (193.2) (351.1)Intercept 1.89 100,080 1,138,117 -25,573 -31,317

(10.1) (119,827) (88,691) (53,128) (101,641)

Number of Obs. 1,807 2,392 175 343R2 0.517 0.101Log-likelihood -18,009.50 -1,433.10Sources: ESS 1992, ESE 1992, DMMO 1992.Notes: Columns (1 ) and (3) give least squares estimates; models (2) and (4) are estimated bymaximum-likelihood (generalized tobit). See Appendix Table A2 for the coefficients of the probit selectionequations. Columns (1 ) and (3) use only establishments with strictly positive costs and strictly positiveterminations for economic reasons. Columns (1 ) and (2) give estimates for establishments with 50 or moreemployees. Columns (3) and (4) give estimates for establishments with 20-49 employees.

Table 4a: The Cost of Terminations in 1992


34/45

Coef.

Independent Variable (St.Err.)Collective Terminations 51,924

(3,632)Collective Terminations (squared) -21.2

(3.84)Intercept for Collective Terminations 1,517,208

(180,949)Individual Terminations -63,361

(143,493)Individual Terminations (squared) 4,003

(15,700)Intercept for Individual Terminations 1,668,254

(267,983)

Number of Obs. 2,392Log-likelihood -18,060.50Sources: ESS 1992, ESE 1992, DMMO 1992.

Notes : Model estimated by maximum-likelihood (generalizedtobit) on establishments with 50 or more employees.

Table 4b: The Cost of Terminations in 1992(distinguishing collective and individual terminations)


35/45

Coef.Independent Variable (St. Err.)

Collective Terminations Eng., Prof., Managers 190,549(36,299)

Collective Terminations Eng., Prof., Managers (squared) -611.3(179.0)

Intercept for Collective Terminations (Eng., Prof., Man.) 765,883(1,178,722)

Collective Terminations Other Skills 24,646(4,042)

Collective Terminations Other Skills (squared) 1.86(5.11)

Intercept for Collective Terminations (Other Skills) 1,147,132(865,668)

Individual Terminations Eng., Prof., Managers -193,014(162,096)

Individual Terminations Eng., Prof., Managers (squared) 46,645(22,380)

Intercept for Individual Terminations (Eng., Prof., Man.) 112,967(847,105)

Individual Terminations Other Skills -24,992(134,020)

Individual Terminations Other Skills (squared) -2,075(14,120)

Intercept for Individual Terminations (Other Skills) 1,238,190(861,563)

Number of Observations 1,365

Log-likelihood -10,282.80Sources: ESS 1992, ESE 1992, DMMO 1992.Notes :Model estimated by maximum-likelihood (generalized tobit) onestablishments with 50 or more employees. See Appendix Table A2 for thecoefficients of the probit selection equations.

Table 5: The Cost of Terminations by Skill-Levels in 1992


36/45

Coef.Independent Variable (St. Err.)

Collective Terminations 76,768(5,133)

Collective Terminations (squared) -14.7(5.89)

Intercept for Collective Terminations 5,065,036(2,612,723)

Individual Terminations 39,946(224,692)

Individual Terminations (squared) -1425.3(24,206)

Intercept for Individual Terminations 4,875,678(2,623,745)

Total Employment 760,497(248,048)

Total Employment (squared) 0.2 10 -5

(2.0 10 -5)Firm in Expansion -253,337

(258,540)Firm in Contraction 304,610(300,092)

Within-Firm Seniority Distribution (first decile) 10,852(47,246)

Within-Firm Seniority Distribution (first quartile) -46,298(48,711)

Within-Firm Seniority Distribution (median) 26,428(36,599)

Within-Firm Seniority Distribution (third quartile) -4,949(36,610)

Within-Firm Seniority Distribution (ninth decile) 43,106(27,096)

Within-Firm Wage Distribution (first decile) 236,910(450,680)

Within-Firm Wage Distribution (first quartile) 185,695(421,063)

Within-Firm Wage Distribution (median) 131,848(270,502)

Within-Firm Wage Distribution (third quartile) -268,735(316,103)

Within-Firm Wage Distribution (ninth decile) -851,607(296,576)

Table 6: The Cost of Terminations: decomposing the fixed costs in 1992


37/45


38/45

Variable Coef. St.Err Coef. St.Err Coef. St. Err (1) (2) (3)

Probit (selection):Seniority 0.076 0.012 0.080 0.012Age-Seniority 0.061 0.011 0.065 0.010Managers 0.056 0.189 -0.147 0.195Technicians -0.054 0.181 -0.144 0.187Clerks -0.155 0.188 -0.232 0.194Blue-Collar Workers -0.273 0.176 -0.341 0.182French -0.113 0.113 -0.085 0.120Male 0.183 0.077 0.151 0.082Log wage -0.406 0.054 -0.419 0.053Residual 1.632 0.093 1.623 0.107

Tobit (retirement bonus):Intercept 77,554 9,167 294,023 17,192 12,011 4,298Wage (in 10 6 FFr) x Seniority 5,446 753 1,081 1,811 7,121 720Seniority -2,718 684Age-Seniority -3,181 490Manager 45,932 17,998

TechnicianClerk -11,203 19,824Blue-Collar Worker -17,682 19,275Male 11,117 7,702French -8,087 11,876 68,738 3,396 73,908 4,068Correlation -0.644 0.055 -0.793 0.057

Number of Observations 2,636 2,636 404

Log-Likelihood (or R2) -5,968.80 -5,958.60 0.20

Sources: ESS 1992.Notes: Estimated by maximum likelihood (generalized tobit) for columns (1) and (2); OLS for column (3).

Table 8: Retirement Costs and Individual Characteristics


39/45

Mean Coef. Coef.Variable (St.Dev) (St.Err) (St.Err)

(1) (2)Hiring Costs 218,475 dep. dep.

(2,065,041)Total Hiring 84.7 2,909 2,015

(159.9) (700.5) (780.9)Total Hiring (squared) 32,725 -1.94 -1.42

(182,935) (0.61) (0.65)Intercept 2.10 35,653 385,364

(9.88) (67,193) (137,351)

Number of Obs. 1,562 1,679R 2 0.012Log-likelihood -13,464.80Sources: ESS 1992, ESE 1992, DMMO 1992.Notes: Model (1) uses least squares estimates on the establishments withstrictly positive costs and strictly positive hires. Model (2) is based onmaximum-likelihood estimates (generalized tobit).

Table 9: The Cost of Hiring in 1992


40/45

Mean Coef.Variable (St.Dev) (St.Err)

Hiring Costs 227,171 dep.(2,434,239)

Total Hiring Eng., Prof., Manag. 8.34 68,174(23.06) (10,033)

Total Hiring Eng., Prof., Manag. (squared) 601.0 -188.4(6341) (37.7)

Intercept Eng., Prof., Manag. 2.17 -100,043(10.89) (342,521)

Total Hiring Other Skills 73.76 -1,769(156.6) (1,485)

Total Hiring Other Skills (squared) 29,944 -0.55(184,662) (1.13)

Intercept Other Skills 123.2 -33,044(1,856) (361,702)

Number of Obs. 939Log-likelihood -13,710.10Sources: ESS 1992, ESE 1992, DMMO 1992.Notes: Maximum likelihood estimates (generalized tobit).

Table 10: The Cost of Hiring by Skill-Levels in 1992


41/45

Mean Coef.Variable (St. Dev.) (St. Err.)

Hiring Costs 1,438,353 dep.(8,653,507)Total Hiring, CDI, Eng.,Prof.,Manag. 15.44 270,051

(21.42) (94,536)Total Hiring, CDI, Eng., Prof., Manag. (squared) 691.6 -1,251

(2,677) (743)Total Hiring, CDD, Eng., Prof., Manag. 6.29 82,833

(27.56) (184,684)Total Hiring, CDD, Eng., Prof., Manag. (squared) 790.0 -344.6

(6,784) (748.0)Total Hiring, CDI, Other Skills 29.64 75,669

(60.84) (56,890)Total Hiring, CDI, Other Skills (squared) 4,536 -168.9

(23,649) (136.3)Total Hiring, CDD, Other Skills 127.8 -16,054

(238.1) (12,788)Total Hiring, CDD, Other Skills (squared) 72,378 -9.21

(230,965) (12.35)Intercept 5.06 -2,224,813

(12.16) (1,581,474)

Number of Observations 86R 2 0.187Sources: ESS 1992, ESE 1992, DMMO 1992.Notes: Least squares estimates. The models use only those establishments withstrictly positive costs, with strictly positive hires, and reported data on hires by

types of contract and skill-levels.

Table 11: The Cost of Hiring by Skill-Level and Contract Type in 1992


42/45

Estimated Number of

Workers

Expected Completed

Job Duration

Average Annual

Wage

Annualized Adjustment

Cost

(Adjustment Cost)/

(Total Compensation)Workers less than age 50 Managers, Engineers and Professionals 11.0 6.50 362,115 16,629 4.4%Middle-level Managers and Professionals 29.6 6.92 204,489 17,699 8.0%Clerks 45.3 5.33 144,149 13,651 8.7%Unskilled White Collar Workers 9.4 3.08 123,429 7,878 6.0%Skilled Blue Collar Workers 48.9 6.07 144,514 15,531 9.7%Unskilled Blue Collar Workers 34.1 4.68 121,898 11,971 8.9%

Workers age 50 and over

Managers, Engineers and Professionals 2.0 8.84 465,645 13,181 2.8%Middle-level Managers and Professionals 3.4 10.01 246,270 14,923 5.7%Clerks 3.6 9.09 166,095 13,554 7.5%Unskilled White Collar Workers 0.8 7.24 124,748 10,796 8.0%Skilled Blue Collar Workers 5.0 9.76 156,652 14,546 8.5%Unskilled Blue Collar Workers 2.7 8.88 131,461 13,229 9.1%Source: DADS 1976-1996.

Table 12: Estimated Annualized Effect of Adjustment Costs on Per Worker Total Labor Cost


43/45

Employment GrowthCategory

Hired intoLongTerm

Contract

Hired intoShortTerm

ContractTotalEntry Quits

End of ShortTerm

Contract Retirement TerminationsTotalExits

Establishments with 9.8 26.9 37.2 9.6 17.9 0.6 1.9 30.3increasing employment (83.4) (187.6) (201.4) (62.2) (158.3) (6.9) (21.4) (187.6)in year t (all years)

(N=3,465)

Establishments with 5.1 17.4 22.7 8.8 16.2 1.2 2.7 29.8decreasing employment (48.1) (160.5) (170.9) (53.8) (154.0) (12.7) (31.0) (178.5)in year t (all years)

(N=3,179)

Establishments with 7.1 15.5 22.7 8.5 12.1 0.8 1.6 23.1stable employment (61.9) (128.6) (138.6) (65.1) (114.1) (9.6) (13.6) (140.9)in year t (all years)

(N=371)Sources: Abowd, Corbel and Kramarz (1999) based on Table 8. Original source: DMMO, 1987-1990;weighted by ex post weights.

Appendix Table A1: Rates of Entry/Exit of Workers by Employment Growth Category andType of Employment Contract or Separation (per 100 Employees)


44/45

Appendix Table A2: Selection Probits to Accompany Retirement and Termination Equations (Text Tables 2-5)

Retirement Equations Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.

Trainees' Compensation (young) -0.0404 0.0166 0.0787 0.0553 -0.0603 0.0323Trainees' Compensation (others) 0.0423 0.0109 0.0834 0.0316 -0.0098 0.0255Training Costs per Worker 0.0160 0.0080 0.0565 0.0321 -0.0065 0.0294Human Resources Dept. (professionals) -0.0012 0.0029 -0.0917 0.2029 0.0679 0.0486Human Resource Dept. (technicians) 0.0012 0.0033 0.1129 0.1719 0.0209 0.0287Human Resource Dept. (clerks) 0.0121 0.0026 0.0131 0.1369 0.0146 0.0289Total Employment 0.4330 0.0424 8.7644 6.7001 0.0003 0.0002Training Spending (professionals) -0.5410 0.2441 -0.1206 1.2220 -0.5441 0.3501Training Spending (others) -0.9582 0.7109 -0.3911 1.0259 0.0689 1.6686

Average Labor Costs (per worker) 0.0024 0.0004 -0.0001 0.0005 0.0100 0.0013Good Business Conditions 0.0031 0.0449 -0.0538 0.1522 0.0063 0.0851Share of Professionals, Technicians -2.2016 0.6951Share of Clerks 0.5094 0.4578Share of Blue-Collar Workers 0.4444 0.4107Constant -0.5195 0.0662 -1.0743 0.3099 -1.4563 0.4515

Termination Equations Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.Trainees' Compensation (young) -0.0163 0.0198 -0.0155 0.0542 -0.0430 0.0243Trainees' Compensation (others) -0.0112 0.0121 0.0235 0.0365 0.0248 0.0236Training Costs per Worker -0.0212 0.0100 -0.0856 0.0305 0.0224 0.0171Human Resources Dept. (professionals) 0.0640 0.0199 0.0681 0.2093 0.2733 0.0864Human Resource Dept. (technicians) -0.0160 0.0062 0.0123 0.1128 0.1045 0.0225Human Resource Dept. (clerks) -0.0097 0.0053 -0.0811 0.1372 -0.0744 0.0090Total Employment 1.1919 0.1210 -3.6085 6.7408 2.8376 0.2887Training Spending (professionals) 0.6647 0.4541 -2.3556 1.4194 2.7709 0.1670Training Spending (others) -1.2733 0.6948 -1.9869 1.6502 -9.6009 0.0630Average Labor Costs (per worker) 0.0009 0.0006 0.0012 0.0006 0.0020 0.0011Good Business Conditions 0.5745 0.0617 0.6096 0.1559 0.5528 0.0859Share of Professionals, Technicians 2.6412 0.7094Share of Clerks -0.2820 0.3240Share of

Documents

Abowd Kramarz Costs Hiring Separation Final